A method and system for assisted reproductive management

Abstract

The application relates to the technical field of assisted reproductive health management, and discloses an assisted reproductive management method and system. The method comprises the following steps: receiving and standardizing physiological data of a plurality of monitoring devices, performing time sequence alignment and feature fusion to form a multi-dimensional feature matrix; utilizing pattern clustering and a sliding window technology to identify the dynamic evolution track of a key physiological pattern; performing state deduction on the pattern sequence based on an individualized model to predict a future physiological state; generating a regulation strategy sequence according to the difference between the prediction and a target, and executing an instruction; comparing real-time physiological response data with an expected model to dynamically adjust a subsequent strategy, thereby forming a closed-loop optimization; and finally evaluating the effectiveness of a path according to complete cycle data and updating model parameters. The application realizes continuous and accurate prediction of a reproductive physiological state and individualized adaptive regulation, and improves the accuracy and efficiency of management.

Description

Technical Field

[0001] This invention relates to the field of assisted reproductive health management technology, specifically to an assisted reproductive management method and system. Background Technology

[0002] Current assisted reproductive technology (ART) management primarily relies on discrete medical tests and static clinical indicator assessments. A common practice is for patients to regularly visit medical institutions for ultrasound examinations and hormone level blood tests. Doctors, based on the results of a single or limited number of tests, combined with their personal experience, determine the patient's menstrual cycle stage and formulate subsequent drug intervention or treatment plans. This management model treats the continuous physiological process as a series of isolated snapshots, resulting in information that is both delayed and incomplete.

[0003] Existing technological solutions have shortcomings. Data acquisition is intermittent, failing to capture continuous dynamic changes in physiological parameters within a cycle, resulting in limited accuracy in predicting key physiological events. Treatment strategy formulation and adjustment rely on retrospective judgments at discrete time points, lacking analysis of the dynamic evolution of individual physiological states. Once a treatment plan is determined, it typically remains fixed throughout the treatment cycle, unable to be quickly and accurately adjusted based on the patient's real-time response to medication. The entire management process is highly dependent on physicians' clinical experience, making it difficult to ensure consistency in decision-making, and patients require frequent hospital visits, placing a heavy burden on them.

[0004] Currently, assisted reproductive technology (ART) management lacks an intelligent system capable of integrating multi-source continuous data, automatically identifying the dynamic trajectory of individual physiological patterns, and forming closed-loop regulation based on real-time feedback. How to achieve continuous and accurate perception and prediction of reproductive physiological status, and how to construct a regulatory mechanism that can adapt to individual responses and dynamically optimize treatment strategies, are key issues that need to be addressed to improve the success rate and efficiency of ART. Summary of the Invention

[0005] The purpose of this invention is to provide an assisted reproductive technology management method and system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for assisted reproductive technology management, the method comprising:

[0007] The system receives reproductive physiological data streams from multiple monitoring devices, performs data cleaning and outlier correction operations on the data streams to generate standardized physiological parameter sequences, and performs time-series alignment and feature fusion processing on the standardized physiological parameter sequences to form a multidimensional physiological feature matrix.

[0008] Pattern clustering analysis is performed on the multidimensional physiological feature matrix to identify key physiological patterns related to the reproductive cycle; the dynamic evolution trajectory of the key physiological patterns is extracted based on a sliding time window to generate a pattern evolution sequence.

[0009] A personalized physiological state model is established, the pattern evolution sequence is input into the personalized physiological state model to perform state inference, and the physiological state prediction results at future time points are output.

[0010] Based on the difference between the predicted physiological state and the preset reproductive target state, the state regulation quantity is calculated; a regulation strategy sequence is generated based on historical regulation records and the state regulation quantity.

[0011] The regulatory instructions in the regulatory strategy sequence are executed, and physiological response data after regulation are collected in real time. The physiological response data is compared with the expected response model, and the subsequent regulatory strategy sequence is dynamically adjusted.

[0012] The effectiveness of the reproductive state migration pathway is evaluated based on the physiological response data sequence within the complete regulatory cycle; the parameters of the personalized physiological state model are updated based on the evaluation results.

[0013] Preferably, the step of performing time-series alignment and feature fusion processing on the standardized physiological parameter sequence includes:

[0014] Key physiological event markers in the standardized physiological parameter sequence are detected, and multi-source physiological parameters are normalized over time based on the key physiological event markers.

[0015] The time-domain features, frequency-domain features, and nonlinear features of the normalized physiological parameters are extracted, and the multidimensional physiological feature matrix is ​​generated by a feature weighted fusion algorithm.

[0016] The multidimensional physiological feature matrix is ​​subjected to dimensionality reduction processing, retaining feature dimensions that are more correlated with reproductive status than a preset threshold.

[0017] Preferably, the identification of key physiological patterns related to the reproductive cycle includes:

[0018] Density clustering algorithm is used to perform cluster analysis on the feature points in the multidimensional physiological feature matrix to identify high-density feature regions;

[0019] Calculate the center point and distribution range of each of the high-density feature regions, and define the center point as a typical physiological state;

[0020] Analyze the transition probabilities between the typical physiological states and construct a physiological state transition network;

[0021] State paths related to the progress of the reproductive cycle are extracted from the physiological state transition network and used as the key physiological patterns.

[0022] Preferably, the step of extracting the dynamic evolution trajectory of the key physiological pattern based on a sliding time window includes:

[0023] A variable-length sliding time window is set and continuously slid over the standardized physiological parameter sequence;

[0024] Within each sliding time window, the pattern strength index and pattern stability index of the key physiological pattern are calculated;

[0025] Record the change curves of the model strength index and model stability index over time to form the model evolution sequence.

[0026] Preferably, the establishment of a personalized physiological state model includes:

[0027] Collect patients' historical reproductive physiological data and extract the coupling relationships and nonlinear interactions between physiological parameters;

[0028] A state deducer based on a neural network is constructed, wherein the network structure and connection weights of the state deducer are initialized according to the coupling relationship and nonlinear interaction;

[0029] The state inferr is trained using a reinforcement learning algorithm so that it can accurately predict the evolution trend of physiological states.

[0030] Verify the prediction accuracy of the state inference device. When the prediction error is lower than a preset threshold, confirm that the personalized physiological state model has been established.

[0031] Preferably, the calculated state control quantity includes:

[0032] Define a distance metric function between the current physiological state and the target reproductive state in a multidimensional feature space;

[0033] The value of the distance metric function is calculated, and the basic regulatory quantity is obtained by combining it with the constraint of the rate of change of physiological state;

[0034] Query the control effect data of similar states in the historical control records, and correct the basic control amount;

[0035] Taking into account individual patient differences, the modified control quantity is adjusted in a personalized manner to generate the final state control quantity;

[0036] The process of personalizing the modified control quantity by incorporating individual patient differences to generate the final state control quantity includes:

[0037] Obtain a dataset of individual variability factors for patients, including age, body mass index, basal metabolic rate, expression levels of genetic markers, and past reproductive history.

[0038] A mapping model between individual difference factors and the sensitivity of response to regulation was established. The mapping model was trained by analyzing the correlation between individual difference factors and regulation effects in historical patient data.

[0039] The corrected control amount is input into the mapping relationship model to calculate the control amount adjustment coefficient;

[0040] The modified control amount is scaled according to the control amount adjustment coefficient to obtain the preliminary personalized control amount.

[0041] A real-time physiological parameter feedback mechanism is introduced to monitor the patient's current stress level and metabolic status indicators, and to dynamically fine-tune the initial personalized regulation amount.

[0042] The final state control quantity is generated by combining the results of the scaling process and dynamic fine-tuning.

[0043] Preferably, the step of generating a control strategy sequence based on historical control records and the state control quantity includes:

[0044] The state control quantity is decomposed into step-by-step control targets for multiple time stages;

[0045] For each step of the control objective, the most suitable control measure is matched from the control strategy library;

[0046] Arrange the timing and intensity gradient of the aforementioned control measures to form a preliminary control strategy sequence;

[0047] Simulate the execution of the initial regulatory strategy sequence to predict possible fluctuations in physiological responses;

[0048] The preliminary control strategy sequence is optimized and adjusted based on the prediction results to generate an executable control strategy sequence.

[0049] The step of optimizing and adjusting the initial control strategy sequence based on the prediction results to generate an executable control strategy sequence includes:

[0050] The fluctuation amplitude in the physiological response fluctuation prediction report is compared with the preset safety threshold to identify the control period that needs to be optimized;

[0051] For each control period that needs optimization, the timing or intensity gradient of the control measures is adjusted to generate multiple candidate control strategy sequences.

[0052] The simulation process was re-executed for each candidate regulatory strategy sequence to evaluate the fluctuation level of the optimized physiological response.

[0053] The candidate regulatory strategy sequence that minimizes the fluctuation level of physiological response is selected as the optimized regulatory strategy sequence.

[0054] Verify whether the optimized regulatory strategy sequence conforms to the path constraints for achieving reproductive target states;

[0055] Finally, an executable sequence of control strategies is generated, and the optimization and adjustment records of each control measure are marked.

[0056] Preferably, comparing the physiological response data with the expected response model includes:

[0057] Establish an expected response model that includes the normal response range, the over-response range, and the under-response range;

[0058] The deviation between the actual physiological response data and the normal response range in the expected response model is calculated in real time.

[0059] When the deviation exceeds the preset tolerance, the control strategy adjustment mechanism is triggered;

[0060] Based on the direction and magnitude of the deviation, determine the adjustment direction and magnitude of the control strategy.

[0061] Preferably, the effectiveness of assessing the migration path of reproductive status includes:

[0062] Compare the degree of agreement between the actual reproductive migration path and the expected migration path;

[0063] Analyze the causes of path deviation and distinguish between systematic deviation and random fluctuation;

[0064] The calculation path migration efficiency metrics include migration time, energy consumption index, and stability index;

[0065] Based on the aforementioned consistency and path migration efficiency indicators, a path effectiveness evaluation report is generated.

[0066] Preferably, when the processor executes the computer program, it implements the steps of the assisted reproductive management method as described in any one of the above-described methods.

[0067] Compared with the prior art, the beneficial effects of the present invention are:

[0068] By receiving continuous data streams from multiple monitoring devices and performing time-series alignment and feature fusion, a multidimensional physiological feature matrix is ​​formed. Then, pattern clustering and sliding window analysis are used to identify the dynamic evolution trajectory of key physiological patterns. This method shifts reproductive health management from relying on static, discrete parameter indicators to analyzing continuous, dynamic physiological pattern sequences. State extrapolation based on pattern evolution sequences enables earlier and more accurate prediction of future physiological states. This dynamic pattern-based prediction provides a more precise time window for intervention.

[0069] By establishing a personalized physiological state model and generating a sequence of regulatory strategies based on the difference between the predicted results and the target, and by collecting physiological response data in real time after the instructions are executed and comparing it with the expected model to dynamically adjust subsequent strategies, a complete "perception-decision-execution-feedback" closed loop is constructed. The system can adjust subsequent intervention plans in real time and automatically based on the individual's actual biological response to the preceding regulatory measures. This adaptive mechanism changes the traditional open-loop management model of fixed treatment plans, long adjustment cycles, and reliance on doctors' experience and trial and error. It enables the management process to accurately match the individual's real-time physiological state changes, reduce unnecessary external interventions and their side effects, and improve the personalization and safety of treatment. The entire regulatory process realizes a transformation from passive and lagging adjustment to proactive and forward-looking optimization. Attached Figure Description

[0070] Figure 1 This is a schematic diagram illustrating the working principle of the assisted reproductive management method described in this invention.

[0071] Figure 2 A flowchart for key physiological pattern recognition;

[0072] Figure 3 A flowchart for extracting dynamic evolution trajectories;

[0073] Figure 4 The curves showing the dynamic changes in reproductive physiological parameters during the implementation of the regulation strategy;

[0074] Figure 5 This is a graph showing the trend of estrogen level regulation during follicular development. Detailed Implementation

[0075] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] Please see Figure 1This invention provides a method for assisted reproductive technology management. The method includes: inputting a data stream into the system in the form of a time series, which may contain missing values, noise, or outlier readings. Therefore, a data cleaning operation is first performed, including filling missing values ​​using interpolation methods, smoothing noise using filtering algorithms, and identifying outliers using statistical outlier detection technology. Outliers are then corrected based on the trends of neighboring data points, thereby generating a standardized physiological parameter sequence to ensure data consistency and reliability. Subsequently, the standardized physiological parameter sequence undergoes time-series alignment and feature fusion processing. Time-series alignment involves synchronizing data from different sources on the time axis, using key physiological events such as ovulation markers as a benchmark. Feature fusion extracts multi-dimensional features from the aligned data, including time-domain, frequency-domain, and nonlinear features, and forms a multi-dimensional physiological feature matrix through weighted combination.

[0077] Pattern clustering analysis is performed on the multidimensional physiological feature matrix. Clustering algorithms are used to group data points and identify typical physiological states corresponding to high-density regions. These states are related to different stages of the reproductive cycle, such as the follicular phase, ovulation phase, and luteal phase, thereby extracting key physiological patterns. The dynamic evolution trajectory of these key physiological patterns is extracted based on a sliding time window. The sliding window moves across the time series with a variable length, and pattern indices such as intensity or stability are calculated within each window to generate a pattern evolution sequence reflecting the dynamic changes in physiological states. A personalized physiological state model is established. This model, trained based on historical patient data, may employ neural networks or state-space models. It can accept the pattern evolution sequence as input, perform state extrapolation, and output predictions of physiological states at future time points, such as predictions of ovulation windows or hormone level trends.

[0078] Based on the difference between the predicted physiological state and the preset reproductive target state, a state regulation quantity is calculated. The difference can be quantified using a distance function, and the regulation quantity is optimized considering historical regulation effects and individual differences. A regulation strategy sequence is generated based on historical regulation records and state regulation quantities. Historical records provide references for similar cases. The regulation sequence is decomposed into step-by-step goals and matched with specific regulation methods such as drug dosage or lifestyle adjustments. The regulation instructions in the regulation strategy sequence are executed through automated devices or under physician guidance, while real-time collection of post-regulation physiological response data, such as hormonal changes or physiological indicator feedback. The physiological response data is compared with the expected response model, which defines normal, excessive, or insufficient response ranges. The comparison results are used to dynamically adjust subsequent regulation strategies to achieve closed-loop control. Based on the physiological response data sequence within the complete regulation cycle, the effectiveness of the reproductive state migration path is evaluated. Evaluation indicators include path fit and efficiency. Finally, the parameters of the personalized physiological state model are updated based on the evaluation results to improve future prediction accuracy.

[0079] Example 1: See Figure 2In practical implementation, detecting key physiological event markers in a standardized physiological parameter sequence is a fundamental step. Key physiological event markers include, but are not limited to, luteinizing hormone (LH) peaks, basal body temperature spikes, or dominant follicle rupture signals monitored by ultrasound. The detection process relies on real-time analysis of the physiological parameter sequence. For example, a gradient change in hormone levels is calculated using a sliding window. When the gradient value exceeds a preset threshold, it is marked as a potential event point. Subsequently, an event verification algorithm is applied to eliminate false events caused by measurement noise. The event verification algorithm checks the continuity of data points before and after the point and its synergy with other parameter events, ultimately determining the precise timestamp of the key physiological event marker. In some embodiments, the detection of key physiological event markers integrates multi-source information for cross-validation. For example, the LH concentration curve output by the hormone analyzer is compared synchronously with the basal body temperature curve recorded by the thermometer. Only when data from two independent sources indicate the presence of a significant physiological event within the same time interval is the key physiological event marker finally confirmed. This method enhances the robustness of event detection and avoids misjudgments caused by the failure or anomaly of a single data source.

[0080] In practice, time-axis normalization is performed on multi-source physiological parameters using key physiological event markers as a benchmark. This normalization aims to address the issues of incomplete synchronization of timestamps from different monitoring devices and differences in individual menstrual cycle lengths. The process first aligns the key physiological event markers to the origin of a standard time axis; for example, the detected ovulation day is defined as day 0 of the cycle. Then, the time series data of all physiological parameters are resampled and interpolated around this origin, ensuring that each parameter sequence has the same temporal resolution and length on the normalized time axis. It can be understood that time-axis normalization may employ linear interpolation or spline interpolation algorithms to fill the resampled data points, ensuring the continuity of data over time. The normalized time axis is typically set to a standard cycle length. For individuals with cycle lengths significantly longer or shorter than the standard length, their time axis is compressed or stretched proportionally, while maintaining the position of the key physiological event markers relative to the origin. This process allows data from different individuals and different cycles to be compared and integrated within a unified time frame.

[0081] In specific implementations, the time-domain, frequency-domain, and nonlinear features of normalized physiological parameters are extracted. Time-domain features include mean, variance, standard deviation, skewness, kurtosis, and first and second-order differences, which are directly calculated from the amplitude information of the time series and used to describe the basic statistical characteristics of physiological parameters in the time dimension. Frequency-domain features are obtained by performing Fast Fourier Transform or Wavelet Transform on the time series, extracting the main frequency components, power spectral density, and energy distribution in different frequency bands to reveal the periodic fluctuation patterns of physiological signals. Nonlinear features include approximate entropy, sample entropy, Lyapunov exponent, and fractal dimension, which are used to characterize the inherent complexity and chaotic properties of the physiological system. In some embodiments, the feature extraction process adjusts parameters according to the type and physiological significance of the physiological parameters. For example, for hormone concentration curves, the focus is on their pulsatile secretion characteristics, thereby calculating time-domain and frequency-domain indices such as pulse amplitude and pulse interval; for body temperature curves, more attention may be paid to their circadian rhythm-related frequency-domain features and nonlinear features characterizing the stability of body temperature regulation. The extracted multi-dimensional features form a high-dimensional feature vector, which prepares for subsequent feature fusion.

[0082] In its implementation, a multidimensional physiological feature matrix is ​​generated using a feature-weighted fusion algorithm. The core of this algorithm lies in assigning appropriate weights to each extracted feature. The weights reflect the importance of the feature in distinguishing different reproductive physiological states. Weights can be determined based on the correlation coefficient between the feature and known reproductive state labels, or by calculating feature importance scores using machine learning algorithms. During execution, the algorithm linearly weights and sums all features corresponding to each normalized time point according to their weights, ultimately forming a weighted feature value. Arranging these weighted feature values ​​in chronological order constitutes one row of the multidimensional physiological feature matrix. Multiple rows obtained from the same processing of multiple physiological parameters together form the complete matrix. Optionally, the algorithm standardizes all features before weighting to eliminate the influence of differences in feature dimensions and numerical ranges, ensuring fairness in the weighting process. The generated multidimensional physiological feature matrix has rows representing time points and columns representing the fused weighted features. This matrix encapsulates the core information of multiple physiological parameters.

[0083] In practice, the multidimensional physiological feature matrix undergoes dimensionality reduction, retaining features whose correlation with reproductive status exceeds a preset threshold. The main purpose of dimensionality reduction is to reduce data complexity, eliminate redundant information, and improve the efficiency of subsequent pattern clustering analysis. Commonly used dimensionality reduction methods include principal component analysis (PCA) and linear discriminant analysis (LDA). PCA projects the original high-dimensional features onto a new set of orthogonal features through linear transformation, sorts them according to their variance contribution rate from largest to smallest, and then retains the top few principal components whose cumulative contribution rate exceeds a preset threshold as the dimensionality-reduced features. LDA is a supervised dimensionality reduction method that attempts to find the projection direction that maximizes the inter-class variance and minimizes the intra-class variance after projecting different categories of data. It is understandable that the selection of the preset threshold requires a trade-off between information retention and model complexity; a threshold that is too high may lead to the loss of useful information, while a threshold that is too low will result in insignificant dimensionality reduction. The dimensionality reduction process ultimately produces a new feature matrix with lower dimensionality but higher information density, which is used for subsequent clustering analysis.

[0084] In practical implementation, density clustering algorithms are used to cluster feature points in the multidimensional physiological feature matrix. One such algorithm is DBSCAN, which can discover clusters of arbitrary shapes without requiring a pre-specified number of clusters. DBSCAN is based on two core parameters: the neighborhood radius Eps and the minimum number of points MinPts. The algorithm starts with any unvisited feature point, finds all points within its Eps neighborhood, and if the number of points in the neighborhood is greater than or equal to MinPts, the point is marked as a core point and a new cluster is created. Then, it recursively visits all points within the neighborhood of the core point, expanding the cluster's range until it can no longer be expanded. The algorithm then continues processing the next unvisited point. Through density clustering, densely distributed points in the multidimensional feature space can be grouped into the same cluster, while points in low-density areas are marked as noise points, thus identifying high-density feature regions. These high-density feature regions correspond to periods of relatively stable or frequently occurring physiological states.

[0085] In practical implementation, the center point and distribution range of each high-density feature region are calculated. The center point is usually defined as the mean vector of all feature points in the cluster, representing the typical location of the cluster. The distribution range describes the dispersion of points within the cluster, which can be characterized by calculating the covariance matrix of the cluster. The eigenvalues ​​and eigenvectors of the covariance matrix indicate the direction and range of the cluster's extension in space. The center point is defined as a typical physiological state. Each typical physiological state corresponds to a stage with clear physiological significance. For example, typical physiological state A may represent the early follicular phase, typical physiological state B may represent the ovulation period, and typical physiological state C may represent the mid-luteal phase. The determination of typical physiological states provides the basis for state transition analysis. The transition probability between typical physiological states is analyzed, and a physiological state transition network is constructed. The calculation of the transition probability is based on historical time series data. The number of transitions between typical physiological states at all adjacent time points is counted. Then, the number of transitions is divided by the total number of transitions starting from that state to obtain the transition probability. For example, the probability P(B|A) of transitioning from typical physiological state A to typical physiological state B is equal to the number of times B occurs after A divided by the sum of the number of times any state occurs after A. The physiological state transition network is a directed graph where nodes represent typical physiological states and directed edges represent possible transitions between states. The weight of each edge is the calculated transition probability. This network intuitively describes the possible paths of reproductive physiological state evolution.

[0086] Example 2: See Figure 3 In practice, setting a variable-length sliding time window is the first step in extracting dynamic evolution trajectories. The length of the sliding time window is not fixed but adaptively adjusted according to the dynamic characteristics of the key physiological patterns themselves. For example, when the system detects that the physiological parameter sequence is in a phase of rapid change, it will automatically adopt a shorter window length to capture the rapid evolution details of the pattern more precisely; while when the physiological parameter sequence is in a relatively stable phase, a longer window length will be used to obtain more stable pattern index estimates and reduce the impact of random fluctuations. The sliding step size of the window can also be flexibly set, using a fixed step size or a variable step size related to key physiological events. For example, the sliding frequency can be increased near the expected ovulation time window. This variable-length sliding time window design allows the analysis process to better match the non-stationary characteristics of reproductive physiological processes.

[0087] In practice, a sliding window is continuously applied across a standardized physiological parameter sequence. The sliding process begins at the start of the time series and moves sequentially forward according to a set step size, ensuring coverage of the complete data range from the beginning to the end of the cycle. For each window position, a segment of the standardized physiological parameter sequence within that window is extracted as the current data block for analysis. Each data block contains the values ​​of multiple physiological parameters within the window's time range. It's understandable that if the data at the ends of the time series is insufficient, preventing the window from fully covering the data, boundary processing is required. Common processing methods include symmetrically expanding the data or directly ignoring incomplete windows to ensure that each data block participating in the analysis has a complete time length. The continuous sliding window process generates a series of temporally continuous or overlapping data blocks, providing input for subsequent calculations within each window.

[0088] In specific implementation, within each sliding time window, the pattern strength index and pattern stability index of the key physiological pattern are calculated. The pattern strength index quantifies the significance of the key physiological pattern in the current window data. Its calculation depends on the definition of the key physiological pattern. If the key physiological pattern is defined as a typical feature vector, the pattern strength index can be obtained by calculating the projection or cosine similarity between the feature vector of the data points in the current window and that typical feature vector. The pattern stability index is used to evaluate the fluctuation of the key physiological pattern within the current window, which can be achieved by calculating the variance of the feature vector within the window or the coefficient of variation of the pattern strength index within the window. In some embodiments, the pattern strength index... A concrete example of this calculation can be represented as the average correlation coefficient between data points within the window and the pattern prototype, which is formally described as follows:

[0089]

[0090] in: Representative in window The model intensity index obtained from internal calculation, It is a window The total number of data points contained therein This represents the function for calculating the Pearson correlation coefficient. It is a window Inner Multidimensional physiological feature vectors corresponding to each time point It is the prototype vector of key physiological patterns.

[0091] In practice, the curves of model strength and stability indices over time are recorded to form a model evolution sequence. For each sliding time window, after calculating the corresponding model strength and stability indices, these two index values ​​are associated with the center or starting time point of that window and stored chronologically. The model strength index changes over time to form a model strength evolution sequence, and the model stability index changes over time to form a model stability evolution sequence. Together, they constitute the model evolution sequence. Essentially, the model evolution sequence is a time series curve, with time on the horizontal axis and index values ​​on the vertical axis. This curve visually demonstrates how key physiological models evolve from weak to strong and then back to weak throughout the observation period, and how their stability fluctuates. The model evolution sequence captures the continuous characteristics of physiological state evolution, providing direct input data for subsequent personalized physiological state models. Optionally, before forming the final model evolution sequence, the initially calculated index sequence can be smoothed and filtered to suppress potential short-term noise interference, thereby more clearly showing the long-term evolution trend of the model.

[0092] In practice, establishing a personalized physiological state model requires collecting the patient's historical reproductive physiological data. This data should cover multiple complete reproductive cycles as much as possible, including standardized physiological parameter sequences obtained from multiple monitoring devices, pattern evolution sequences extracted through the aforementioned process, and known reproductive event markers. Extracting the coupling relationships and nonlinear interactions between physiological parameters is fundamental to model construction. Coupling relationships can be analyzed by calculating cross-correlation functions, time-delay cross-correlation, or Granger causality between different physiological parameter time series, thereby determining the lead-lag relationships and directions of influence between parameters over time. Detecting nonlinear interactions may require more complex methods, such as nonlinear correlation analysis based on mutual information, or reconstructed phase space techniques in nonlinear time series analysis to reveal complex dynamic dependencies between parameters.

[0093] In practical implementation, a state inferrer based on a neural network is constructed. The choice of network structure for the state inferrer needs to fully consider the characteristics of time-series data prediction. Recurrent neural networks or their variants, such as Long Short-Term Memory networks or gated recurrent units, are commonly chosen because these network structures have internal memory mechanisms that can capture long-term temporal dependencies. The input layer dimension of the network corresponds to the feature dimension of the pattern evolution sequence, while the output layer dimension corresponds to the physiological state dimension that needs to be predicted at future time points. The connection weights of the state inferrer are initialized based on the extracted coupling relationships and nonlinear interactions. This is an important pre-training strategy. For example, if the analysis finds that parameter A has a strong unidirectional influence on parameter B, then the initial values ​​of the weights connecting the input node of parameter A to the state prediction node of parameter B can be set relatively large, rather than simply using random initialization. This initialization based on prior knowledge helps to accelerate network convergence and improve the interpretability of the model.

[0094] In practice, reinforcement learning algorithms are used to train the state inferrer. The reinforcement learning framework models the state inference task as an interaction between an agent and its environment: the agent is the state inferrer, and the environment is the simulated physiological state evolution process. At each time step, the agent selects an action based on the currently observed pattern evolution sequence, i.e., predicts the physiological state at the next time step. The environment then provides a reward value, which is calculated based on the difference between the predicted state and the actual state. The training objective is to maximize the cumulative reward, i.e., minimize the long-term prediction error. It is understandable that commonly used reinforcement learning algorithms such as deep Q-networks, policy gradient methods, or actor-critic algorithms can be applied to this training process. Through a large number of iterative training rounds, the state inferrer gradually learns the state transition dynamics hidden behind the pattern evolution sequence.

[0095] In practical implementation, the prediction accuracy of the state extrapolator is verified. When the prediction error is lower than a preset threshold, the personalized physiological state model is considered successfully established. The verification process needs to be conducted on an independent test dataset, which consists of historical patient data not used during training. Prediction error is typically measured using metrics such as root mean square error (RMSE) and mean absolute percentage error (MASE). The preset threshold needs to be set according to the specific clinical application scenario and prediction goals. For example, for ovulation day prediction, the threshold might be set to an error of no more than one day; for hormone level prediction, the threshold might be set to a relative error of no more than 15%. Only when all prediction error metrics of the state extrapolator on the test set are lower than the preset threshold is the personalized physiological state model considered sufficiently reliable and ready for use in practical assisted reproductive technology (ART) management decision support.

[0096] Example 3: In specific implementation, the distance metric function between the current physiological state and the target reproductive state in the multidimensional feature space is defined as the starting point for calculating the state regulation quantity. The current physiological state is represented by the prediction result output by the personalized physiological state model, and is a vector containing multiple physiological parameter values. The target reproductive state is an ideal state vector preset according to the treatment goal. For example, the target state may be set as a combination of estrogen levels, follicle diameters, and endometrial thickness. Distance metric function We can use the weighted Euclidean distance method, the specific expression of which is as follows:

[0097]

[0098] in: This represents the calculated distance between the current physiological state and the target reproductive state. The total number of dimensions representing the physiological state vector. The current physiological state vector is at the th position. Component values ​​on the dimension, The target reproductive state vector is in the th order. Component values ​​on the dimension, It is to give the first Weight coefficients of dimensional features, weight coefficients The magnitude of this dimension reflects the importance of this feature in reproductive status assessment and is usually determined by domain knowledge or historical data analysis.

[0099] In practice, the value of the distance metric function is calculated, and combined with the constraint of the rate of change of physiological state, the baseline regulatory quantity and the value of the distance metric function are obtained. It is a scalar value that directly reflects the magnitude of state differences. The constraint on the rate of change of physiological state refers to the limitation on the maximum allowable range of change in physiological state per unit time. This constraint is to avoid discomfort or risk to the individual due to excessively rapid physiological changes; for example, the daily rate of change of hormone levels should not exceed a certain safe threshold. The calculation of basal regulatory quantities needs to consider distance values ​​simultaneously. And the rate of change of the current physiological state; if the current rate of change is already close to the upper limit, even if the distance value... If the current state is relatively large, the basic regulatory amount will be controlled at a relatively moderate level; conversely, if the current state is stable and the rate of change is low, it can be based on the distance value. A relatively large baseline adjustment amount is calculated proportionally. This baseline adjustment amount can be considered as a theoretical initial intervention level.

[0100] In practice, the baseline regulatory level is adjusted by querying historical regulatory records of similar states. The historical regulatory database stores data from past cases, including the physiological state before regulation, the applied regulatory level, and the resulting changes in physiological state. The query process first searches the historical database for past instances most similar to the current physiological state. Similarity is determined based on the aforementioned distance metric function or a specialized similarity metric algorithm. After finding similar cases, the actual effects of applying different regulatory levels in these cases are analyzed, including target achievement rate and side effect incidence. The adjustment logic is as follows: if historical data shows that a smaller regulatory level can achieve good results under similar states, the baseline regulatory level is adjusted downwards; if historical data shows that stronger intervention is needed, it is adjusted upwards. This historical experience-based adjustment makes the baseline regulatory level more closely reflect actual conditions.

[0101] In practice, the modified regulatory amount is personalized by incorporating individual patient differences to generate the final state regulation amount. A dataset of individual patient differences is acquired. This dataset consists of static and semi-static parameters collected before or during treatment, including age, body mass index, basal metabolic rate, expression levels of reproductive-related genetic markers, and detailed past reproductive history. This data is obtained through methods such as consultation, physical examination, laboratory testing, and gene sequencing, and is structured and stored in electronic health records. A mapping model is established between individual patient differences and the sensitivity of the regulatory amount's response. This mapping model is essentially a regression or classification model; its input is a vector of individual patient differences, and its output is a predicted value of the sensitivity of the regulatory amount's response. Response sensitivity can be quantified as the change in physiological state caused by a unit of regulatory amount. Establishing this mapping model requires analyzing the correlation between individual patient differences and regulatory effects in historical patient data. The training process uses a large amount of historical patient data on individual patient differences as features and their actual regulatory effect data as labels, employing machine learning algorithms to learn the complex mapping relationship from individual features to response sensitivity.

[0102] In practice, the corrected regulatory amount is input into the mapping model to calculate the adjustment coefficient. The corrected regulatory amount serves as an input parameter to the mapping model, and is input along with the individual difference factor dataset into the pre-trained model. The mapping model predicts the patient's expected response sensitivity to the regulatory amount based on the input individual difference factors. This predicted sensitivity is then compared to a preset baseline sensitivity; the ratio, or a scaling factor calculated based on that ratio, is the adjustment coefficient. The adjustment coefficient is a dimensionless value. If it is greater than 1, it indicates that the patient's sensitivity to the regulatory amount is above average, and the regulatory amount may need to be reduced; if it is less than 1, it indicates low sensitivity, and the regulatory amount may need to be increased.

[0103] In practice, the modified regulatory amount is scaled based on the adjustment coefficient to obtain the initial personalized regulatory amount. This scaling typically uses multiplication, meaning the initial personalized regulatory amount equals the modified regulatory amount multiplied by the adjustment coefficient. The direction of the scaling process depends on the definition of the adjustment coefficient and the specific implementation of the mapping model. The core principle is to ensure that the final regulatory amount produces similar expected physiological effects in the specific patient as in the general population, thus achieving truly individualized drug administration or intervention. A real-time physiological parameter feedback mechanism is introduced to monitor the patient's current stress level and metabolic status indicators, dynamically fine-tuning the initial personalized regulatory amount. This mechanism involves continuously monitoring key dynamic physiological parameters of the patient during the execution of the regulatory strategy. Stress level indicators can be assessed using heart rate variability, cortisol salivary concentration, or subjective fatigue scale scores; metabolic status indicators can include real-time blood glucose levels, respiratory quotient, or dynamic energy expenditure. These indicators are obtained through wearable devices, rapid testing reagents, or periodic assessments. The dynamic fine-tuning process is a closed-loop control process. The system adjusts the initial personalized regulatory amount by small increments or decrements based on the deviation between the real-time monitored indicator values ​​and the expected range. For example, if a patient's stress level is detected to be significantly elevated, indicating that the patient may be in a state of over-stress, the system may automatically reduce the medication dosage by a small percentage temporarily for safety reasons.

[0104] Example 4: In practical implementation, the state regulation quantity is decomposed into step-by-step regulation targets for multiple time stages. The state regulation quantity is a comprehensive total intervention amount, which needs to be rationally allocated to different time stages of a complete reproductive cycle. The division of time stages follows the natural laws of reproductive physiology and usually includes early follicular phase, late follicular phase, ovulation phase, early luteal phase, and late luteal phase. Each step-by-step regulation target represents a specific range of physiological indicators to be achieved within a specific time stage. For example, in early follicular phase, the target might be to slowly increase the follicle diameter to a specific size, while in ovulation phase, the target is to precisely trigger follicle rupture. The decomposition process needs to consider the temporal distribution logic of the state regulation quantity. For example, in the first half of the cycle, a gradually increasing intervention might be used, reaching a peak near the critical event, and then gradually weakening. The step-by-step regulation targets constitute a target sequence, providing a clear direction for subsequent matching of specific regulation methods.

[0105] In practice, referring to Table 1, for each step-by-step regulatory goal, the most suitable regulatory approach is matched from the regulatory strategy library. The regulatory strategy library is a structured knowledge base or database storing various validated intervention methods. Each record in the library is associated with its applicable physiological stage, expected physiological parameters, standard dosage or intensity range, and known precautions. The matching process is based on the core requirements of the step-by-step regulatory goal, querying and filtering within the regulatory strategy library. For example, for the goal of "promoting follicle growth," possible regulatory approaches matched from the library might include "using gonadotropin injections," "supplementing with specific doses of vitamin D," and "implementing specific dietary recommendations." The matching algorithm comprehensively considers the fit between the goal and the approach, the safety record of the approach, and the patient's past tolerance, recommending one or more candidate regulatory approaches for each step-by-step regulatory goal.

[0106] Table 1: Fragment Table of Regulation Strategy Library

[0107]

[0108] In practical implementation, the timing and intensity gradient of regulatory measures are arranged to form a preliminary regulatory strategy sequence. After determining the regulatory measures required for each stage, they need to be arranged in chronological order, and the starting time and duration of each measure are set. The intensity gradient refers to the change in the intensity of the same regulatory measure at different time points. For example, a drug may start at a low dose and gradually increase the dose as the cycle progresses, forming a "step-like" or "gradually increasing" intensity gradient to avoid initial overstimulation. Therefore, the preliminary regulatory strategy sequence is an ordered list containing three core elements: time, means, and intensity, detailing each step of the intervention plan from the beginning to the end of the cycle. It is understandable that arranging the timing and intensity gradient needs to follow pharmacokinetic principles and physiological rhythms; for example, some drugs need to be taken at fixed times to maintain stable blood drug concentrations, and some physical therapies need to be avoided during rest periods.

[0109] In practice, the initial regulatory strategy sequence is simulated to predict potential fluctuations in physiological responses. This simulation relies on a built-in physiological response model that characterizes the effects of different regulatory measures applied to specific physiological states. Using the initial regulatory strategy sequence and the patient's current physiological state as input, the physiological response model calculates predicted values ​​for various physiological parameters at each future time point. By analyzing the trajectory of these predicted values, potential fluctuations in physiological responses can be identified, such as sudden increases or decreases in hormone levels, or excessively rapid or slow follicle growth. The predictive model can be based on a population pharmacokinetic / pharmacodynamic model or a personalized model built using the patient's historical data. The simulation process serves as a "rehearsal" before the actual intervention, assessing the safety and effectiveness of the strategy.

[0110] In practice, the initial control strategy sequence is optimized and adjusted based on the prediction results to generate an executable control strategy sequence. The fluctuation amplitude in the physiological response fluctuation prediction report is compared with a preset safety threshold to identify control periods requiring optimization. The physiological response fluctuation prediction report is a direct output of the simulation execution, quantifying the predicted amplitude and rate of change of key physiological parameters. The preset safety threshold is a boundary value set based on clinical guidelines and expert experience to define a safe range of physiological fluctuations. The predicted fluctuation amplitude at each time point is compared with the corresponding safety threshold; any time period where the fluctuation amplitude exceeds the safety threshold is marked as a control period requiring optimization. These periods typically indicate overstimulation, under-response, or other potential risks.

[0111] In practice, for each regulation period requiring optimization, the timing or intensity gradient of the regulatory measures is adjusted, generating multiple candidate regulation strategy sequences. For each identified regulation period requiring optimization, it is analyzed which regulation measures(s) caused the drastic fluctuations in the physiological response. Adjustment methods include, but are not limited to: advancing or delaying the start time of the regulatory measures, decreasing or increasing the intensity parameter of the measures, or replacing it with a milder equivalent measure from the regulation strategy library. Each adjustment to one or more parts of the initial regulation strategy sequence generates a new candidate regulation strategy sequence. Therefore, for a complex initial scheme, multiple different candidate regulation strategy sequences may be generated for further evaluation.

[0112] In practice, the optimized regulatory strategy sequence must be verified to ensure it conforms to the path constraints for achieving reproductive goals. These constraints refer to essential physiological laws and clinical rules that must be followed. For example, follicles must have sufficient growth time to trigger ovulation, and the endometrium must reach a certain thickness before embryo transfer. Besides minimizing fluctuations, the optimized regulatory strategy sequence must also meet these rigid constraints. The verification process involves checking whether the key event time points and the sequence of physiological state transitions in the sequence meet these constraints. If not, the sequence needs to be readjusted or candidate sequences need to be reselected until a scheme that minimizes fluctuations and meets all path constraints is found.

[0113] In practice, an executable regulatory strategy sequence is generated, with each regulatory measure's optimization and adjustment records annotated. After screening and verification through the above steps, an optimal regulatory strategy sequence is determined. This executable regulatory strategy sequence is output in a structured format (such as XML or JSON), clearly listing the execution time, specific means, intensity parameters, and duration of each regulatory step. Simultaneously, the sequence includes detailed metadata, annotating the optimization and adjustment records of each regulatory measure compared to the initial preliminary regulatory strategy sequence, such as "delaying the gonadotropin initiation date by 2 days" or "reducing the ovulation-inducing hormone dose by 500 IU." These records provide clinicians with transparent decision-making traceability for reviewing protocols and facilitate subsequent analysis and experience accumulation. Optionally, the executable regulatory strategy sequence can be directly integrated into a medical information system to generate treatment schedules or guide automated drug delivery devices.

[0114] See Figure 4This visually presents the dynamic evolution of follicle diameter, estrogen levels, and progesterone levels during the implementation of assisted reproductive technology (ART) strategies. Specifically, the curves, with cycle days as the horizontal axis, simultaneously demonstrate the changing trends of these three physiological parameters: follicle diameter (blue solid line) shows a phased increase from the early to late follicular phase, reaching a peak of approximately 22 mm in the late follicular phase and then stabilizing; estrogen levels (red dashed line, unit × 10 pg / mL) rise rapidly during the follicular phase, reaching a peak of 400 pg / mL (corresponding to 40 in the figure) before ovulation, and then gradually decline as the cycle progresses; progesterone levels (green dashed line, unit ng / mL) remain low during the follicular phase, rise significantly after ovulation, and approach 14 ng / mL in the late cycle. The dynamic relationship of these three parameters reflects the core logic of reproductive regulation: the increase in follicle diameter is highly synchronized with the increase in estrogen levels, reflecting the effectiveness of follicle-stimulating measures; while the late rise in progesterone levels corresponds to the implementation of luteal phase support measures, and together with the decrease in estrogen levels, constitutes the physiological basis for endometrial receptivity. The parameter changes at different stages in the figure (such as the rapid increase in follicle diameter in the late follicular phase and the significant increase in progesterone in the luteal phase) are highly consistent with the expected mechanisms of action of the corresponding regulatory measures in the regulatory strategy library (such as urinary gonadotropin injection in the late follicular phase and progesterone supplementation in the luteal phase), which intuitively verifies the execution effect of the regulatory strategy sequence.

[0115] Example 5: In practical implementation, establishing an expected response model that includes normal response range, excessive response range, and insufficient response range is the basis for comparison. The expected response model is a set of standard response patterns predefined for a specific regulatory strategy and a specific patient group. The normal response range describes the ideal range of physiological parameter changes expected under the action of the regulatory strategy, such as the reasonable rate of daily increase in estrogen levels after injection of ovulation-inducing drugs. The excessive response range defines situations where physiological parameter changes exceed the safe upper limit, such as early indicators of ovarian hyperstimulation syndrome. The insufficient response range refers to situations where physiological parameter changes do not reach the expected lower limit, such as delayed follicular development. The expected response model can be represented as the boundary of a region in a multidimensional space, or it can be a set of rules about the trajectory of parameter changes. The establishment of the model relies on the statistical analysis of a large amount of historical clinical data and the summary of domain expert knowledge.

[0116] In practice, the deviation between the actual physiological response data and the normal response range in the expected response model is calculated in real time. The actual physiological response data consists of physiological parameter values ​​collected in real time by monitoring equipment during the execution of the regulatory strategy sequence, such as daily serum hormone concentrations and follicle size. Calculating the deviation requires comparing the actual data at each time point with the corresponding normal response range in the expected response model to determine the degree of deviation. Here is an example:

[0117]

[0118] in: It represents the scalar value of the overall deviation calculated at a specific point in time. This represents the number of physiological parameters being monitored. Representing the The actual measured values ​​of each physiological parameter. The first in the expected response model The center value of the normal response range of each physiological parameter at the corresponding time point. The first in the expected response model The width of the normal response range for each physiological parameter.

[0119] In practice, when the deviation exceeds a preset tolerance, a control strategy adjustment mechanism is triggered. The preset tolerance is a threshold set based on clinical safety and treatment goals. The deviation is calculated in real time. If the value consistently exceeds this threshold, or if a single calculated value significantly exceeds the threshold, the system will automatically trigger a regulatory strategy adjustment mechanism. This triggering mechanism can be a software interrupt signal alerting the clinician to intervene and review, or it can be the system automatically initiating a strategy adjustment algorithm. In some embodiments, the preset tolerance may not be a single fixed value, but a dynamic threshold related to the treatment stage. For example, during the critical period approaching ovulation, the tolerance range may be narrowed to implement more precise control. Based on the direction and magnitude of deviation, the adjustment direction and magnitude of the regulatory strategy are determined. The deviation direction is determined by the position of each physiological parameter's actual value relative to the center value of the normal response range, and can be categorized as positive or negative deviation. The deviation amount is the specific numerical value of each parameter's deviation from the center value. The logic for determining the adjustment direction is: if the overall deviation direction points to over-response, the adjustment direction of the regulatory strategy should be to weaken the intervention intensity; if it points to under-response, the adjustment direction should be to strengthen the intervention intensity. The determination of the adjustment magnitude is related to the overall degree of deviation. The individual deviations of each parameter are proportional and usually follow a preset control rule, such as the idea of ​​a proportional-integral-derivative controller, so that the adjustment range can quickly correct the deviation without causing violent oscillations in the system.

[0120] In practice, the degree of agreement between the actual reproductive state migration path and the expected migration path is compared. The actual reproductive state migration path is the state evolution trajectory reconstructed in the physiological state space based on the continuous collection of physiological response data sequences throughout the entire regulation cycle. The expected migration path is the ideal state evolution trajectory predicted at the beginning of treatment based on a personalized physiological state model and a pre-set regulation strategy. Comparing the degree of agreement requires aligning and comparing the two paths in the time-state space. The minimum cumulative distance between paths can be measured using a dynamic time warping algorithm, or by calculating the average of the cosine similarity sequences of the state vectors of two paths at different time points. A high degree of similarity indicates that the actual physiological evolution largely follows the expected plan, while a low degree of similarity suggests unexpected reactions or external interference. Analyzing the causes of path deviations involves distinguishing between systematic biases and random fluctuations. Path deviations refer to the differences between the actual reproductive migration path and the expected migration path. Analyzing the causes requires a comprehensive judgment combining records of the regulatory process, individual patient data, and external environmental factors. Systematic biases typically manifest as continuous, unidirectional deviations, such as unknown metabolic differences in a patient's response to a certain drug, leading to consistently lower or higher-than-expected drug efficacy. Random fluctuations, on the other hand, manifest as irregular, transient deviations, possibly caused by measurement errors, changes in the patient's daily diet and lifestyle, or other accidental factors. Distinguishing methods include performing time-series analysis on the deviation sequence to check for trends or periodicity, or conducting hypothesis testing to determine if it originates from a random distribution with a mean of zero.

[0121] In practical implementation, path migration efficiency indicators are calculated, including migration time, energy consumption index, and stability index. Migration time refers to the actual time taken for the reproductive state migration path to reach the target state from the initial state, compared with the expected time. The energy consumption index is a metaphorical indicator used to quantify the "resources" consumed to achieve the target state. In the context of assisted reproduction, it can be defined as a comprehensive measure of the total drug dosage used throughout the entire regulation cycle, the total number of medical procedures performed on the patient, or the total cost of treatment. The stability index is used to assess the smoothness of the migration path, which can be measured by calculating the variance of the first and second differences of the state vector. The smaller the variance, the smoother the path and the higher the stability. These efficiency indicators evaluate the quality of the regulation process from different dimensions. Based on the concordance and path migration efficiency indicators, a path effectiveness assessment report is generated. The path effectiveness assessment report is a structured document or electronic record that systematically summarizes the overall performance of the reproductive state migration path during this regulation cycle. The report will include the concordance. The report includes quantitative scores, qualitative analysis of the causes of path deviation, and specific values ​​and comparisons with expected values ​​for efficiency indicators such as migration time, energy consumption index, and stability index. The report concludes with an overall evaluation of the effectiveness of the control strategy sequence used, highlighting successes and areas for improvement. This report serves as a crucial basis for updating the parameters of the personalized physiological state model and provides valuable empirical data for the treatment of similar patients in the future.

[0122] See Figure 5In the analysis of estrogen level response during the follicular development phase of the regulation cycle, a reference system was constructed with the number of regulation days on the horizontal axis and estrogen level on the vertical axis. This system included the expected response center value, the normal response range, the over-response threshold, and the under-response threshold. The dynamic trajectory of the actual response value (red line) needs to be compared with the expected system: in the initial stage, the actual value rises rapidly and is within the normal response range; as regulation progresses, the actual value shows a fluctuating increase within the normal response range, and gradually approaches the over-response threshold in the later stages. The core value of this graph lies in its intuitive presentation of the matching relationship between the actual estrogen level response and the expected model, providing a visual basis for judging the degree of response deviation and triggering adjustments to the regulation strategy. Its data presentation method conforms to the quantitative analysis standards for physiological response monitoring in assisted reproductive technology management.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for managing assisted reproduction, characterized in that, The method includes: The system receives reproductive physiological data streams from multiple monitoring devices, performs data cleaning and outlier correction operations on the data streams to generate standardized physiological parameter sequences, and performs time-series alignment and feature fusion processing on the standardized physiological parameter sequences to form a multidimensional physiological feature matrix. Pattern clustering analysis is performed on the multidimensional physiological feature matrix to identify key physiological patterns related to the reproductive cycle; the dynamic evolution trajectory of the key physiological patterns is extracted based on a sliding time window to generate a pattern evolution sequence. A personalized physiological state model is established, the pattern evolution sequence is input into the personalized physiological state model to perform state inference, and the physiological state prediction results at future time points are output. Based on the difference between the predicted physiological state and the preset reproductive target state, the state regulation quantity is calculated; a regulation strategy sequence is generated based on historical regulation records and the state regulation quantity. The regulatory instructions in the regulatory strategy sequence are executed, and physiological response data after regulation are collected in real time. The physiological response data is compared with the expected response model, and the subsequent regulatory strategy sequence is dynamically adjusted. The effectiveness of the reproductive status migration pathway is evaluated based on the physiological response data sequence within the complete regulatory cycle; the parameters of the personalized physiological status model are updated based on the evaluation results. The identified key physiological patterns related to the reproductive cycle include: Density clustering algorithm is used to perform cluster analysis on the feature points in the multidimensional physiological feature matrix to identify high-density feature regions; The center point and distribution range of each high-density feature region are calculated, and the center point is defined as a typical physiological state. Each typical physiological state corresponds to a stage with clear physiological significance. The transition probability between the typical physiological states is analyzed, and a physiological state transition network is constructed. The calculation of the transition probability is based on historical time series data. The number of transitions between typical physiological states at all adjacent time points is counted, and then the number of transitions is divided by the total number of transitions starting from that state to obtain the transition probability. The physiological state transition network is a directed graph, where nodes are various typical physiological states, directed edges represent possible transitions between states, and the weight of the edge is the calculated transition probability. This network intuitively describes the possible paths of reproductive physiological state evolution. State paths related to the progress of the reproductive cycle are extracted from the physiological state transition network and used as the key physiological patterns.

2. The assisted reproductive management method according to claim 1, characterized in that, The time-series alignment and feature fusion processing of the standardized physiological parameter sequence includes: Key physiological event markers in the standardized physiological parameter sequence are detected, and multi-source physiological parameters are normalized over time based on the key physiological event markers. The time-domain features, frequency-domain features, and nonlinear features of the normalized physiological parameters are extracted, and the multidimensional physiological feature matrix is ​​generated by a feature weighted fusion algorithm. The multidimensional physiological feature matrix is ​​subjected to dimensionality reduction processing, retaining feature dimensions that are more correlated with reproductive status than a preset threshold.

3. The assisted reproductive management method according to claim 2, characterized in that, The extraction of the dynamic evolution trajectory of the key physiological patterns based on a sliding time window includes: A variable-length sliding time window is set and continuously slid over the standardized physiological parameter sequence; Within each sliding time window, the pattern strength index and pattern stability index of the key physiological pattern are calculated; Record the change curves of the model strength index and model stability index over time to form the model evolution sequence.

4. The assisted reproductive management method according to claim 3, characterized in that, The establishment of a personalized physiological state model includes: Collect patients' historical reproductive physiological data and extract the coupling relationships and nonlinear interactions between physiological parameters; A state deducer based on a neural network is constructed, wherein the network structure and connection weights of the state deducer are initialized according to the coupling relationship and nonlinear interaction; The state inferr is trained using a reinforcement learning algorithm so that it can accurately predict the evolution trend of physiological states. Verify the prediction accuracy of the state inference device. When the prediction error is lower than a preset threshold, confirm that the personalized physiological state model has been established.

5. The assisted reproductive management method according to claim 4, characterized in that, The calculated state control parameters include: Define a distance metric function between the current physiological state and the target reproductive state in a multidimensional feature space; The value of the distance metric function is calculated, and the basic regulatory quantity is obtained by combining it with the constraint of the rate of change of physiological state; Query the control effect data of similar states in the historical control records, and correct the basic control amount; Taking into account individual patient differences, the modified control quantity is adjusted in a personalized manner to generate the final state control quantity; The process of personalizing the modified control quantity by incorporating individual patient differences to generate the final state control quantity includes: Obtain a dataset of individual variability factors for patients, including age, body mass index, basal metabolic rate, expression levels of genetic markers, and past reproductive history. A mapping model between individual difference factors and the sensitivity of response to regulation was established. The mapping model was trained by analyzing the correlation between individual difference factors and regulation effects in historical patient data. The corrected control amount is input into the mapping relationship model to calculate the control amount adjustment coefficient; The modified control amount is scaled according to the control amount adjustment coefficient to obtain the preliminary personalized control amount. A real-time physiological parameter feedback mechanism is introduced to monitor the patient's current stress level and metabolic status indicators, and to dynamically fine-tune the initial personalized regulation amount. The final state control quantity is generated by combining the results of the scaling process and dynamic fine-tuning.

6. The assisted reproductive management method according to claim 5, characterized in that, The generation of the control strategy sequence based on historical control records and the state control quantity includes: The state control quantity is decomposed into step-by-step control targets for multiple time stages; For each step of the control objective, the most suitable control measure is matched from the control strategy library; Arrange the timing and intensity gradient of the aforementioned control measures to form a preliminary control strategy sequence; Simulate the execution of the initial regulatory strategy sequence to predict possible fluctuations in physiological responses; The preliminary control strategy sequence is optimized and adjusted based on the prediction results to generate an executable control strategy sequence. The step of optimizing and adjusting the initial control strategy sequence based on the prediction results to generate an executable control strategy sequence includes: The fluctuation amplitude in the physiological response fluctuation prediction report is compared with the preset safety threshold to identify the control period that needs to be optimized; For each control period that needs optimization, the timing or intensity gradient of the control measures is adjusted to generate multiple candidate control strategy sequences. The simulation process was re-executed for each candidate regulatory strategy sequence to evaluate the fluctuation level of the optimized physiological response. The candidate regulatory strategy sequence that minimizes the fluctuation level of physiological response is selected as the optimized regulatory strategy sequence. Verify whether the optimized regulatory strategy sequence conforms to the path constraints for achieving reproductive target states; Finally, an executable sequence of control strategies is generated, and the optimization and adjustment records of each control measure are marked.

7. The assisted reproductive management method according to claim 6, characterized in that, The step of comparing the physiological response data with the expected response model includes: Establish an expected response model that includes the normal response range, the over-response range, and the under-response range; The deviation between the actual physiological response data and the normal response range in the expected response model is calculated in real time. When the deviation exceeds the preset tolerance, the control strategy adjustment mechanism is triggered; Based on the direction and magnitude of the deviation, determine the adjustment direction and magnitude of the control strategy.

8. The assisted reproductive management method according to claim 7, characterized in that, The effectiveness of assessing the migration pathways of reproductive status includes: Compare the degree of agreement between the actual reproductive migration path and the expected migration path; Analyze the causes of path deviation and distinguish between systematic deviation and random fluctuation; The calculation path migration efficiency metrics include migration time, energy consumption index, and stability index; Based on the aforementioned consistency and path migration efficiency indicators, a path effectiveness evaluation report is generated.

9. An assisted reproductive management system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the assisted reproductive management method according to any one of claims 1 to 8.

Images